Page 64 - Read Online

Page 59 Li et al. Intell Robot 2021;1(1):58-83 I http://dx.doi.org/10.20517/ir.2021.08

1. INTRODUCTION
From the first stirrings of life, nature has been providing a suitable breeding ground for the intelligence of
organisms. Biological intelligence enables organisms to adapt the extreme or changing environments. For
instance, a group of birds and fishes can efficiently sense the surrounding dynamic environments and take
effective actions based on those inputs often with very simple mechanisms and with limited availability of
information. Some species exhibit collective behaviors and can cooperatively accomplish tasks that are beyond
the capabilities of a single individual under limited implicit communication. Organisms with such beneficial
traits can pass on these traits to offspring, exhibiting high adaptability to environments. The nervous system
in the brain gives human abilities of feeling, thinking, and learning abilities.

Recently, therehasbeenageneralmovementtowardsservice-orientedrobotsthatrequiretheabilitytoadaptto
complex dynamic situations and to handle various uncertainties. Due to the desirable properties of biological
organisms, such as adaptability, robustness, versatility, and agility, the researchers have been trying to infuse
robots with biological intelligence that will enable safe navigation and efficient cooperation among the au-
[1]
tonomous robots in changing environments . The approaches inspired by biological intelligence are known
[2]
asbiologicallyinspiredintelligence,whichhasbeenexploredandstudiedformanyyearsinroboticsresearch .
The fundamental idea of biologically inspired intelligence is to incorporate useful biological strategies, mecha-
nisms, and structures into the development of new methodologies and technologies to solve existing problems
in a more efficient way than existing methodologies and technologies. For instance, swarm intelligence and
collective behaviors of living organisms have inspired the design of many robotics algorithms based on their
biological strategies [3,4] . The process of natural selection has inspired many computational models to opti-
[7]
mize robot performances, such as genetic algorithm [5,6] and differential evolution . The neural network
algorithm, derived from neural science, has gained rising popularity among researchers around the world [8,9] .
Biologically inspired intelligence algorithms were also integrated with various conventional algorithms to de-
velop more efficient algorithms. For example, a knowledge based genetic algorithm, which incorporated the
domain knowledge into its specialized operators, was proposed to efficiently generate collision-free path of
robots [10] . A neural network was used to convert the improved central pattern generator output to the foot
trajectories of quadruped robots [11] . However, most bio-inspired studies are limited to conceptual or labo-
ratory investigations or do not have much biological inspiration. Thus, the development of new intelligent
strategies, algorithms and technologies are still highly needed, such as real-time collision-free navigation algo-
rithms of individual robots or communication, coordination, and cooperation algorithms for multiple robotic
systems, to accomplish multi-objective tasks in dynamic environments.

Bio-inspired neurodynamics models have been studied for real-time path planning and control of various
[2]
robotic systems during the past decades . The shunting neurodynamics model was derived from Hodgkin
and Huxley’s membrane models for dynamic ion exchanges [12] . Based on the shunting neurodynamics model
and its model variants, several new algorithms have been successfully developed for real-time path planning
and control of various autonomous robots [13,14] . The definition of real-time is in the sense that the robot path
planner and controller respond immediately to the dynamic environment, including the robots, targets, ob-
stacles, sensor noise and disturbances. Many other model variants have been also developed for robot path
planning and control. The additive model is computationally simpler and can generate real-time collision-
free paths under most conditions [13,15] . The gated dipole model shows excellent performance in multi-robotic
path planning and tracking control [16] . Beyond the application of autonomous robots, bio-inspired neuro-
dynamics models have been also widely applied to many other research fields, such as odor dispersion with
electronic nose [17] and dynamic ginseng drying [18] . These researches on agriculture have also been extended
to biomedical and other industrial applications.

This paper focuses a comprehensive survey of the state-of-the-art research on bio-inspired neurodynamics
models with their applications to path planning and control of autonomous robots. A detailed introduction